 I'll be honest, we're not, we're not on a charge or something. And it's hard because, okay, so we give you these cases, but you're not an attorney. We always say the man represents himself as a pool of hurt clients, that's the thing. And it's hard because sometimes these guys can't afford, so I don't wanna make light of that. The honest that this is, is that we're just not focused on that, and we could, right? But we are focused on frankly being a startup company and making revenue points and things like that. But the technology we're building could certainly be made available. So the red's turn is what the person's letting us do. So Brian, as you use AI, in that context where you have a non-well-off person and they want to see what they can do, just type in like this scenario and then have some kind of pop-up show where the case law would be good for them to see if they actually stand a chance and maybe allow them to represent themselves. Imagine Ms. Carter just again, trained by experts in say, domestic violence law given some pro bono hours, donated some pro bono hours to this effort so that a cartridge would begin to reason like domestic violence or given jurisdiction. And someone could ask, what are my legal rights? Do I have any legal rights? And they could just reform, explain their scenario. And the system would understand two different people explain scenario, similar scenario in two different ways. What they mean, what their intent is and then put forth recommendations, yes. Leslie, run up against a wall which is you're not allowed to practice law without a license. And various states have been more or less aggressive about that, I mean, going after like a legal Zoom or something like that. I'm assuming that the law would change to a company at all. Do you guys see one regulation set, AI being used by regulators to use to receive financial services, and other these things like blockchain and AI because... I think that's what you were saying about being able to test law. Well, being able to actually test, to back test regulations and to see if regulations that are in place achieve the results that they're put in place to achieve. Rather than, rather than winging it, rather than thinking based upon looking at one instance or two instances that we need this regulation. A protocol in our first year regulation from a given regulations. But another thing that we're doing is on the regulatory side, there are a whole other slew of what's in these cases for reading the unstructured data, reading the actual regulations, mapping them to control standard. Batman, you better jump in. You better just jump in. Oh, okay. When I was in law school 40 years ago, Lexus ran a new one in Westlaw that didn't exist. Not quite. Westlaw must have existed. It was found in the mid-1812. No, no, West existed. Oh, Westlaw, I was just saying, okay, sorry, leave me the electronic one. Okay, scary. Okay, and when they first came out, they just had keynotes and a full text section because where West, we were founded in 1880. We know the books, right? So they had a bit of a television like most established companies. And for about 15 years, and then there was a business for about 15 years, selling for words. We did fine, but I wouldn't wish it on most people. I was gonna actually give Justice Holmes quote there, but we tried some expert systems back in the 90s. And what is the essence of what you're doing now is different from the failed efforts? I think that Watson must understand, in a sense, understand more natural language. But the law is not the same in every state. The law is changes from Congress to Congress. There is no overriding principle like in biology. There are lots of things that hang together. There are some legal principles come from contradictory sources and they don't agree with each other. In most cases they do, but the hard cases they don't. So could you tell me what's different now versus 25 years? One, there's more data and more access to it, to both machine learning and natural language processing algorithm models by far and away. And the efficacy of those models has increased more. But maybe we're not different. So one of the reasons why we do more business corporate law departments than law firms is kind of what you're touching upon. When we do meet with law firms, we will seek to what is your highest volume area of litigation or which area drives the most revenue for the firm. We'll focus on that, discrete area, and build teach Watson the ontology of that particular practice area, rather than across everything. And then through user acceptance training, the end user will interrogate the system, validate the results. And the system is performing thousands of different calculations behind the scenes and it's making weeks of judgment. And after a while, after the solution is validated, it takes eight weeks or so, you can go live with that. I think is that we're not taking, we're taking a much more discrete approach, much more focused approach than we did in the past. We have access to more data, as I said, and there is more data. And we're able to do more with unstructured data, looking for concepts across documents, even documents that might not have anything on their face to do with each other, to look for non-obvious patterns and trends that otherwise, you have to know what you're looking for beforehand, older machine learning models. Yeah, because expert systems were very critical in some sense. Yeah, and their rules face one plus one equals two with true, with the ability to do a lot with natural language understanding, as we call it. You can potentially preserve your rules and preserve your decision tree, but fit more into that tree because you understand that this is more appropriately slotted here and this is more appropriately slotted here. A much less brittle system. I'd be interested in seeing it. I happen to be listening to it's make-up on the way over there. Skeptical about machines understanding things in the way that humans do. And I don't know that Watson tries to do it the same way, but I think there's a fundamental problem with combination learning. Absolutely. I mean, Kerry doesn't understand anything. Yeah. And the fabric that does unify, the thing we can find 200 years ago was judges citing to precedence, right? And even certain patterns, like the phrase, it is well settled, which is this really unique phrase that means we're not gonna fight about whatever comes after what I'm about to say. And so it's survived not just 100,000 of the time, but you go all the way back to like when the S's were F's and it's like it is well-feddled and that sort of thing. And you see this, no, it's an interesting concept, right? This little four word, three of which are stop words that you can almost throw away, right? Has survived and flourished and gone everywhere. And you can see it. So common law, there are these unifying threads that you can pull out. You're not understanding anything, right? You're just leveraging these patterns that are there. And I think that's a huge amount of power that we've been sort of just leaving on the table. So it's really, is Watson basically pattern-based, what you're saying? Is it a high-falutin pattern-based? In a sense, you want to know what you're looking for. You're not, you know, hey, ask around the question, you know, which is why we try to make sure business is really close to telling. You know generally what you're looking for, even though it might be stated in many different ways. And so you show Watson ground-toothed representatives of the concepts, the themes, the ideas, the goals, the information that you want. This is what duty looks like. This is what breach looks like. This is what causation and damages look like. The elements of the negligence cause of action. And then after a while it will make leaps of judge say, okay, plaintiff met his burden of proof or did not meet his burden of proof, theoretically. But you're building a model, of course. Yeah. Yeah. Anyway, is it possible to say you said something? Just contact me. I'd be happy to give you a technical, anything. And just, and even better, we've got some sort of arrangement as leading up to the event and with legal hackers for some amount of access to some of the APIs. So something that I wanted to offer is part of the sort of weekly hangout that some of us are starting to do just in the Black Couch area in Sandy Pentland's lab as a discussion group. I think we should explore some of the Watson APIs, at least one of those weeks, and sit down and just roll our sleeves up. So not to the exclusion, if like if Brian couldn't offer you like a technical walkthrough, don't miss it. But also we can just hack around a little bit and put some data in and do that. And if we have access to this part of what we can do in the context of a class and many of us would like to learn that. So that's an open offer. So anyone that wants to hack around and some of the APIs that have been made available to us, you know, just say so after class and then we'll make one of these weekly discussion groups. Just focused on that. And then a lot of other hands it grew up. There was a distinction between applying technology to a business versus a practice of law. I'm curious about your guys opinions of where is the law profession most amenable to applying technology to practice of law? That's what you're doing. So I think what you've seen pretty good progress was an e-discovery, right? So this would be, in my day, we get literally boxes. I can't believe I get to say it in my day now, but I guess it's there, right? Or you get literally boxes and boxes of all the emails, right? And you just go through them and apply them, right? Then you started to get, okay, just big, big, you know hard drives with tons of documents and you go one at a time, right? And you'd see things like every Tuesday morning and 8 a.m. it's an email from ESPN with the same subject line. And you'd be like, dude, can I please, I'm just gonna bash you, right? I'm just gonna filter them all out and I'm gonna be like, don't be a cowboy, Pablo. And I'm like, you know. So, and you get back to having a fish in this book, right? You just have terror, I'm just so sucking terrible at that. So now, okay, well they'll say, okay, we'll just review 10,000 documents and the coding of the machine will say, I get it, I get it, if I see these to that word, right? So in the discovery, I've been kind of, actually kind of impressed with lawyers. They kind of took it on, did it. Judges sort of signed off on it. So that's an area where I think they've been, research, they have a ways to go. Research has been a lot, just a lot trickier. Because they've been using West, see they don't use the discovery tools in law school, right? No one teaches you the discovery in law school. So they come out and they don't know, you know. West law they've been using since they were first year law students. So there's really sort of ingrained it. And there are areas where they've been doing tech. A lot to do with M&A due diligence. Oh yeah, transactional. You know what to do with transactional generally. But M&A due diligence has been very, very right for, various AI startups as an area of focus. Upload a contract and we'll tell you if this clause is different than the other. It was machined by Brian's last bullet. What spoke about the meaning of the word. So it really unpacked. So I'll ask this question quickly for both of you. It was at MIT Media Lab in 2014, where I was really impressed with who were having a thoughtful conversation about the ethics of coding. Never really seen that and seemed like they were kind of figuring it out. When I was in law school in South Africa, law was respected, nothing like that, it was a law. Law was respected often as a barrier of protection against the under-proverged or the downtrodden. So law is esteemed in many ways, although it can be used. Do you see any problems and the ethics of what your powerful weapon slash widget is developing in the practice of the law that sits so carefully on the relationships between humans and humans and institutions? I do. One would be judges training expert systems. Those expert systems inherited the biases of those judges on the one hand. But if you think about it another way, those expert systems could be used as merits to reveal to judges, to reveal to us what our biases are from the beginning. And then we can act accordingly. But if we don't do that, and yes, I think that's the first thought that came to my head. I would say absolutely. So you mentioned before there was Westlaw with West, you would use their key topics. So Westlaw over a century had split all the law, all the cases up into different categories. And those categories were basically a prisoner of their taxonomy. That became how lawyers thought about the law. There's a great paper written by Bob Baring on this. So then key search came along and suddenly we're free. We can now just type where we go and we're off and there's problems with that as well. Now here we are saying, let the AI sort of guide you and tell you, if you type a keyword, sure, but we'll let the AI kind of confuse you. So are we re-enprisoning you now? And are what we think the law is, right? It's a real problem. You want like a kayak, you want different AI voices, right? You want sort of hybrid vigor with different folks and the other eyes like that's a ridiculous case. You should read this one, right? Something that sort of diversifies, but hopefully we'll make all the different voices subscribe one at a time. No, I kid. But I think for anything that's the interface between lawyers and the common law, there's a profound responsibility there because that truly becomes then how the lawyer understood and kind of acted. Do you find that the model that case texts use of using these judicial opinions and links between documents works in some areas of laws better than other? So instead of the area of LARNA, I want to talk about research in sort of two modes, right? So there's mode one legal research is sort of the forest. You want to know what's the big governance standard, right? And there the wisdom of the crowd does better, right? So for instance, if I want to invalidate your patent and say this patent is obvious, right? There's this leading case that the Supreme Court has that actually the fact pattern is about a gas pedal and a little electronic sensor on it, right? But even if I'm doing semiconductor research, I have to pay homage to that case because that's sort of multi-factor test or various things, right? Mode one, the links work well, right? But mode two is where you say, I want an exact fact pattern, right? And that's where the wisdom of the crowd actually pulls you away from what you want. Unless you just happen to be doing exactly the same thing that the crowd case has. So I don't think that, I mean, I think that that network of precedent applies to all areas of law. I mean, if you have less case law, you know, certain little niche areas might have very few cases where perhaps they're sort of different factions are. But it's really something that spans all the different topics a lot, sort of how we do it. So I think we have time for one more question for you in about five minutes left. But you get it, I guess. Yeah, I'm wondering if there's a little machine that needs to stop. Do you have a good strategy to make sure the machine's actually doing it? At a certain point, can you stop understanding unless you just like the fact that it's not going to bring to you guys identified lazy, any convenience that you might be willing to extract from the kinds of texts and topics that you're reading, things like what's a mutual way of structuring different causes or subtracting different levels and understanding. So you're talking about sort of black box algorithms where, you know, it's doing something well, but it won't tell you exactly what works. You know, I'm just sort of like, you got to make your piece with that because there's sort of, it's kind of ingrained in it, right? I feel like sometimes it's like, I could tell you if you'd like to read the million words in a list that I've used, but like I'll read the first five and trust you that the rest are kind of... So I guess I'm not quite... Well, yeah, I guess it's a good strategy. One is like trying to find some kind of like truth, I suppose, amongst all the news, I guess, that I haven't seen it, have you? What is truth? Yeah, I know it's never come out. I'm sorry, it didn't go well. It's not... Right. It's not gonna be the same for the way that's interpretable in any way that you would. That sounds like some advanced sort of understanding stuff. I can throw out one thing, which is actors and actions, you know, like the rudiments, the primitive of a use case and law, I would always look for what are the roles of parties and the transactions or in business, you know, something similar. So I've got a lot of mileage, you know, parsing through, I don't know how much data in the law, pulling out scenarios and being able to define the actors and actions and the parties and the transactions. And when you say the role, if it's like, in Uber case, if you say that role is an employee, that tells you a lot. If you say the role is a contractor, it tells you a lot, it's a consumer, tells you a lot, so as soon as you can see a role and start to go on to an actor and what kind of action was it going to work, slipping and falling and whatever it is, that's the closest I've come, but I feel like through the lens of these new algorithms and a lot of data through a social physics lens, we can't begin to understand finally and see the contours of the law the first time. You choose different primitives. So is an Uber driver like a DoorDash guy? Yeah, right, they're both summoned by the app to go and do something, right? But then you get on another one and say, no, that guy's not chauffeuring you around. So what primitive you choose can change depending on your context. So I think efforts to sort of get that tectonic, like this is what this thing is, are always going to be sort of fuzzy if you go sort of out of meetings by the next weird content. Oh guys, this is so fun, I'm looking forward to it. You guys have some really good questions. Thank you, I really love it. So before Yoast says the word, Brian in public, you've given your email addresses, is there a place that you can point, is there an ask you of people that they should understand your thoughts and what you're working on too, is there a place on social media where they can sort of track where you're going to be speaking and what you're thinking about, what you're saying. So what do we actually think about and where to find you and talk to you? I would just say keep up exactly what you guys are doing. Take classes like this, go to these workshops that are being set up. It would be thrilling to have some of the absolutely fantastic firepower of MIT and then you have some of the things that you'd buy. Brought to bear on this tool. As you can see, we need a lot of help and lots. So just keep it up is what I would say. And then follow and say to me, I'm going to tweet it, I'll put it in my handle. So let's spell it out. Brian, for me it's more linked to my name. Can you post your emails on the website? Yeah, if you don't mind, you bet. Thank you so much. All right, thanks. Thank you so much. So that's the end of today's decline session. One important addition, that is to say that you all should have gotten an email from us, from me in particular, Yoast. Talking about what was happening today's session with the background links to our two guest speakers. If you did not get an email, please go to futurelawmit.org and click on the participant link because that's the way to get on the class email list for the upcoming weeks. And then we'll also be talking further about project possibilities and more. Any final questions? Can I add something? Something people, if you got Yoast's email from me, then you haven't done your job. You have to go get it directly from Yoast, from iForward. Through the sort of stellar class system, those are the key words on top. Yes? Okay, good stuff. Thank you to our guests and I'll see you all next week. Okay. Thanks. Futurelawmit.org for the class and for the conference, it's mit.edu forward slash law.